Accelerator Cards

[sim085]

Hi,

I have some questions regarding accelerator card related to the processors and the memory used on them.

Regarding the processors; where the processors of the accelerator cards all the same. For example the processor 030 on the Derringer and Viper and VXL, GVP530 etc where manufactured by the same manufacturer or they where manufactures by the vendors of these cards?

If they are the same then how come these cards come with different memory configurations; some with 8MB others with 32MB while others with 128MB? Shouldn't all these be able to access the same amount of addresses in memory?

Regards,
Sim085

[keropi]

although the cpu is the same, each manufacturer made their own "extra chips" on accelerators.
think of it as motherboards with the same cpu but from different manufacturers: others support 8gb ram - others just 4.  others have 8 sata ports, others have only 4 and so on... \
it is not the cpu that matters the most, the support it gets is important too.

[sim085]

So is it possible to upgrade such accelerator cards to support more memory? For example can the M-Tec be upgraded to 32MB of memory? (maybe using the same solution to upgrade the A500+ to 2MB chipram)!  

[countzero]

no. To upgrade the addressable RAM you need to increase the address space of your board. Solder a net which connects the unused address lines on the 68030 to a RAM controller you need to make your own. There's also the issue of drivers for the OS to recognise that RAM. It's not impossible but definitely out of practical possibility. Get a better accelerator.

[sim085]

(Note that I ask these question to learn the architecture more then because I want to try this out ... I do not have any soldering skills so far).

From my understanding the 030 processor is connected to an (Memory Management Unit) MMU. Now between the processor and MMU there should be all the addresses available to the processor. Therefore the limitation would be between the MMU and the actual Memory chips. Is this right?

Also what do you mean by OS to recognise the RAM? You mean workbench accesses the Memory directly without the use of an MMU?

[zipper]

The MMU in Amiga systems is rarely used - GigaMem is one that uses it and some remappers, but that's it I think.

[alexh]

The 68000 cannot connect directly to the RAM. 68000 bus cycles and DRAM bus cycles are not the same.

An interface chip must be used. These were mainly PAL/PLD chips with the logic written by the board vendor.

Most 32-bit 680x0 chips have the full 32-bit address bus which give access to 4Gbytes of address space.

However to save money very few board designs forward all address lines to the RAM sockets.

It's not a matter of passing all address lines to the RAM. The board does some address decoding also.

While a worthy home-brew project I feel that with the lack of documentation you might never be able to work out how the logic on the board works.

[sim085]

What is confusing me is this; What happened if the processor wanted to store something in an address which was not there? Or this never happened since the processor knew the maximum amount of memory allocated to it?

Also does the operating system talk to these interface chip? or it only talks with the processor? and aren't these interface chips like an MMU?

 

[countzero]

the processor doesn't store anything to a location unless it is given to it by the operating system (in this case, exec.device I believe) you need some kind of program which put the extra RAM into available RAM pool. I think all processor boards have a little ROM in them which contains this, ones that do not, (like derringer) you need to run a program at boot to activate the RAM.

[sim085]

Ok I think I understand (don't do that surprised face please :-)) The operating system would tell the process what instruction to execute and where to store the result and the operating system would know the allowed memory space because of this small program you mentioned. is this right? or I am confusing things?

[countzero]

yes, you get the picture. there are ways to trap the boot process of an amiga and insert these to the available memory pool. Ralph Babel's Guru Book explains in detail, but it's kind of hard to get a legal copy of that book. But you can find *ahem* other *ahem* versions if you know where to look.

[sim085]

Thanks, I will try to search for that book.

... just another question about something which has been confusing me a lot ...

In the Amiga Hardware Database there is written that the A500 can only have a maximum of 32MB fast RAM on processor board (i.e. the accelerator card).

http://amiga.resource.cx/mod/a500.html

That said how come the Viper came with 128MB of fast RAM on the processor board? Is the extra RAM for nothing (which I do not think)? and what determines the limit of 32MB mentioned in Amiga Hardware Database?

http://amiga.resource.cx/exp/ematrix530

I think I asked a similar question about chip ram but I am still not sure if I understand this correctly!

[darksun9210]

sorry for the long winded rambling reply, but it's friday, and i'm hungover (still), and i'm on a single minded misson to avoid work today.

motorola made the 68020, 68030, 68040, and 68060.
different manufacturers took those chips (GVP, Phase5, SSL, M-Tec etc.) and designed boards, and memory controllers to plug those 680x0 processors into amigas.

and in theory, yes, a full 68030 (and 68020) has access to the full 4GB address range no matter who made the card. however, the memory controller is a different matter.
the GVP A530 for example, memory controller maps its memory to the 8MB zorro2 area in the amiga's address range, and so has a limit at 8MB, as the memory controller cannot address any memory outside of that window.
"back in the day" this wasn't really a problem, as 4MB memory modules cost around £150+. so 8MB of ram was not only a lot, but prohibitivly expensive too. plus it didn't take too much controller logic to map the memory to an area the amiga already had set aside for that kind of thing.
this also has the added advantage that even if you switch off the accelerator, it could act as just a ram card for the 68000 chip as the 8MB is well within its addressable area. however that usually didn't happen, as disabling the accelerator disabled the memory controller it had and there for, the memory as well.
the VXL memory controller can take more than 8MB as it maps outside of the 16MB amiga range, into the 32bit space, but can't understand how to talk to memory modules with higher density cell counts above 32MB, as 64 and 128MB simms were again uncommon, and very expensive, and probably developed later than the card was designed.

The A500 Viper530 also mapped its ram above the standard amiga range in the 32bit area, so was happy to take 128MB on one simm slot as it had plenty of room to map it (i suspect that an additional simm slot would be able to take another 128MB simm quite happily..).

The A1200 blizzardIV/PPC series of accelerators have memory controllers that are again able to map to the entire 32bit range, and, understand double banked simm modules up to 128Mb. and as the A1200 doesn't have to worry about Zorro3 space in the 32Bit address range, so the accelerator has almost the whole 4GB range free to map the memory outside of the standard amiga 16/24bit address space and so can take as much ram as you can throw at it. hence 2x simm slots= 256Mb ram. i'm fairly willing to bet that if you could solder another simm socket or two on there, it would happily use those as well.

The A3/4000 series have a slightly different problem. while they are 32bit, and have the whole 4GB at their disposal, their address range has to take account the zorro3 address space, so they only have 128MB assigned to the CPU slot above the standard memory mapping, then the rest is assigned to Zorro3 address space.
so you can have upto 128Mb on the CPU slot, (a crazy amount when the A3000 was conceived), and up to 512MB addressable per zorro3 slot, with 7 slots with available DMA can take up to 3.5GB of ram. but that all goes through the Zorro3 controller (not as fast as ram directly on the CPU card).

But apparently there is a bug in the exec, and expansion.library that means a machine can't see more than 384MB of ram.

when i say standard amiga memory layout, i mean (as far as i can figure):-
starting from the bottom address
----------------------- start of 16bit area
000000000 - 0001FFFFF .. 2MB chipram
000200000 - 0009FFFFF .. 8MB Zorro2 autoconfig space (16bit/gvp A530 ram goes here)
000A00000 - 000BEFFFF .. Reserved
000BFD000 - 000BFDF00 .. CIA-B even addresses only
000BFE001 - 000BFEF00 .. CIA-A odd addresses only
000C00000 - 000D7FFFF .. 512KB slow ram on A500
000D80000 - 000DBFFFF .. Reserved
000DC0000 - 000DCFFFF .. Realtime Clock
000DD0000 - 000E7FFFF .. Reserved
000E80000 - 000E8FFFF .. autoconfig initialization space
000E90000 - 000EFFFFF .. autoconfig space for I/O boards
000F00000 - 000FBFFFF .. 256KB kick rom (high)
000FC0000 - 000FFFFFF .. 256KB kick rom (low)
----------------------- end of 16bit area
001000000 - FFFFFF000 .. 3.5+GB free for use by 32bit accelerators on non zorro3 machines (VXL etc.)
----------------------- end of 32bit area

otherwise, as a rough guess for A3k/4k machines...

----------------------- end of 16bit area
001000000 - 001FFFFFF .. 16MB mainboard fast ram
002000000 - 009FFFFFF .. 128MB CPU slot area
00A000000 - FFFFFF000 .. zorro3 space from the end of the CPU slot to the top of the 32bit range
----------------------- end of 32bit area

i'm sure someone can correct me on these address ranges where i've gone wrong, but the idea is pretty much there

[countzero]

I think they only tried to mean that the accelerator with the most RAM on for A500 had 32 mbs. And they didn't think of viper at the time of writing. There is no such limitation for any amiga (maximum ram on accelerator). A4000's have a 128 mb limit, but that is more related to Zorro3 address space, and any amount above 128mb not being able to DMA by zorro boards, AFAIK.

Chip ram is a completely different thing. It is limited by the chipset and it's inability to address above 2mb. To break this barrier you would need to redesign the whole custom chipset.

[sim085]

From the amiga memory layout there are some things I can understand and some which I cannot (surprise surprise :-))

Always talking from the A500+ perspective (since that is the only machine I have);

'2MB chipram' refers to the 1MB chipram which comes on the motherboard and the 1MB chipram which can be added in the trap door slot.

'8MB Zorro2 autoconfig space' refers to the ram that can be added through for example a side slot expansion such as the GVP A500HD8+ which takes 8MB and the A590 which takes only 2MB if I remember correctly.

'512KB slow ram on A500' I cannot understand ... isnt this the ram that comes on the motherboard and therefore part of the chipram?

'3.5+GB free for use by 32bit accelerators' refers to the memory space that can be on an accelerator card.

Are the above correct; as I said the thing that confused me the most is that 512KB.

The memory table is very helpful to understand the concept. Strange to see that memory wise an A500 can be better equipped (if wanted) then an A4000 :-P